Identifying Dune Habitat through the use of Remote Sensing Classifications




Chavez, Lucas J.

Journal Title

Journal ISSN

Volume Title



The use of object-based image analysis (OBIA) for image classification has grown over the past decade as evidenced by the increasing number of publications across disciplines. In many studies, OBIA is compared with traditional pixel-based image analysis (PBIA), and for some studies, OBIA is reported as producing higher land cover classifications accuracies when compared to PBIA. This study focused on determining whether OBIA classifications of a dune land landscape in West Texas resulted in better accuracy than PBIA classifications. This research used NAIP imagery with 1-m resolution stacked with an NDVI layer as well as inverse distance moment and entropy texture measures for two USGS quadrangles; N Cowden NW and Doodle Bug Well. Supervised PBIA and OBIA classifications were performed using a maximum likelihood classifier to group pixels based on defined classes. For OBIA, an additional step of mean shift segmentation was performed on the imagery to organize pixels into spectrally similar objects. Accuracy assessments were performed for all classifications. Overall accuracies for N Cowden NW PBIA and OBIA were 43.60% (Kappa = 0.2368) and 74.41% (Kappa = 0.5531), respectively. Overall accuracies for Doodle Bug Well PBIA and OBIA were 51.83% (Kappa = 0.2849) and 65.55% (Kappa = 0.4476). To determine whether the difference in classification methods were statistically significant, a test comparing proportions (overall accuracy) was performed to calculate a z-score. Results from the z-tests for comparing PBIA and OBIA for N Cowden NW and Doodle Bug Well were 11 and 5.1. These results indicate that since z ≥ 1.95 (α = 0.05) the overall accuracies for the two classifications are significantly different and that OBIA outperformed PBIA classifications for both study areas.



Remote sensing, Object based, Pixel based, Classification


Chavez, L. J. (2019). <i>Identifying dune habitat through the use of remote sensing classifications</i> (Unpublished thesis). Texas State University, San Marcos, Texas.


Rights Holder

Rights License

Rights URI